【摘要】：隨著互聯(lián)網(wǎng)經(jīng)濟的不斷發(fā)展,提供在線商品和服務(wù)選購的電商平臺的規(guī)模和成交量也在日益增大。這種改變的潮流對在線客服的服務(wù)質(zhì)量和服務(wù)效率提出更高的要求。因此如何通過計算機技術(shù)來輔助人工客服提升其工作效率和工作質(zhì)量是個值得研究的問題。在此基礎(chǔ)上,本文圍繞兩種客服輔助技術(shù)展開:知識庫查詢服務(wù)和客服回復(fù)推薦服務(wù),對相關(guān)的對話模型進行研究和探索。針對客服需要參考相關(guān)專業(yè)知識來完成高質(zhì)量服務(wù)的需求,本文設(shè)計了一種知識庫查詢服務(wù)。該服務(wù)接受用戶的自然語言問句作為輸入,通過AIML模板匹配技術(shù)從輸入中提取關(guān)鍵詞和待查詢屬性,并返回知識庫中對應(yīng)的信息條目。傳統(tǒng)匹配方式受制于自然語言表達的多樣性,存在關(guān)鍵詞匹配失效的問題。本文針對這個問題,提出了一種多輪迭代的同義詞匹配算法,該算法提升了同義詞的檢出數(shù)量和準確度。針對如何提升人工客服工作效率的問題,本文提出了用于客服回復(fù)推薦的深度對話模型。本文從檢索式深度對話模型和產(chǎn)生式對話模型兩個方向來解決該問題。在檢索式的深度對話模型中,本文設(shè)計了一種帶上下文建模的對話模型,通過實驗對比,其比不帶上下文的對話模型有較大性能改善,在此基礎(chǔ)上本文使用用戶咨詢的意圖信息對模型進行了改善,獲得了部分性能提升。在產(chǎn)生式對話模型中,本文設(shè)計了一種使用完整上下文用于預(yù)測客服對話的產(chǎn)生式對話模型,并在客服咨詢數(shù)據(jù)集上同傳統(tǒng)的Seq2Seq模型進行對比,該模型產(chǎn)生的回復(fù)效果更好。在產(chǎn)生式模型中,本文先實現(xiàn)了基礎(chǔ)的Seq2Seq模型,用來作為對照,并根據(jù)本文提出的上下文編碼方式提出并實現(xiàn)了對應(yīng)的產(chǎn)生式對話模型。通過實驗分析,我們發(fā)現(xiàn)本文提出的上下文建模方法對回復(fù)的推薦有提升效果。
[Abstract]:With the development of the Internet economy, the scale and volume of e-commerce platform for online goods and services are increasing day by day. This changing trend demands higher quality and efficiency of online customer service. Therefore, how to improve the efficiency and quality of manual customer service by computer technology is a problem worth studying. On this basis, this paper focuses on two kinds of customer service assistant technology: knowledge base query service and customer service response recommendation service, and studies and explores the relevant dialogue models. A knowledge base query service is designed to meet the needs of customer service which needs to refer to relevant professional knowledge to complete high quality service. The service takes user's natural language questions as input, extracts keywords and attributes from input by AIML template matching technique, and returns corresponding information items in the knowledge base. The traditional matching method is limited by the diversity of natural language expression, and there is the problem of keyword matching failure. In order to solve this problem, a multi-iteration synonym matching algorithm is proposed in this paper, which improves the number and accuracy of synonym detection. Aiming at the problem of how to improve the efficiency of artificial customer service, this paper presents an in-depth dialogue model for customer service response recommendation. This paper deals with this problem from two aspects: the retrieval depth dialogue model and the production dialogue model. In the retrieval model of deep dialogue, this paper designs a kind of dialogue model with context modeling. The experimental results show that the model has better performance than the model without context. On this basis, this paper improves the model by using the intention information of user consultation, and obtains some performance improvements. In the production dialogue model, this paper designs a production dialogue model which uses the complete context to predict the customer service dialogue, and compares it with the traditional Seq2Seq model on the customer service consultation data set. The response effect of the model is better than that of the traditional one. In the production model, we first implement the basic Seq2Seq model, which is used as a contrast, and propose and implement the corresponding production dialogue model according to the context encoding method proposed in this paper. Through experimental analysis, we find that the contextual modeling method proposed in this paper can improve the recommendation of response.
【學(xué)位授予單位】：浙江大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.1

【參考文獻】

相關(guān)碩士學(xué)位論文 前1條

1 朱旺南;基于本體的自動問答客服系統(tǒng)研究[D];青島理工大學(xué);2012年

，

本文編號：2239714